When I upgraded from ambari 2.6 to 2.7(HDP3.1), I uninstalled many services in the cluster for some reason. The end result is ZK Kerberos HDFS.
After I reinstall yarn again. After the yarn-system queue is configured according to the document. On the visit to http://<hostname>:8088, i can actually see the previous task record.
When I looked at the yarn log, I found the yarn resourcemanager that said I had a queue for historical tasks that did not exist. And it refreshes every minute. I don't want to eliminate the error log by adding this useless queue。
I checked the hadoop 3.1 document, found that yarn cli provides a `-destroy ` parameters, but after operation, no work.
I want to know why there will be the previous records after the reconstruction, and how can I clean them up?